Image Segmentation Based on Texture Analysis

OData support
Dr. Pataki Béla
Department of Measurement and Information Systems

In this paper we give a summary of texture analysis methods, and start work on a system that would unify these algorithms and set them up automatically. In this case we take examine one of the most widely used methods, the co-occurrence matrix, and adjusting one of its most important parameters.

Texture analysis is one of the oldest areas of image processing, but many of its fundamental questions are still unclear. Over the years many different algorithms were developed for the analysis of textures, but none of them can be applied universally. In addition, each method needs a complicated set up, so they can only be used by experts.

In the first part of the paper I present the importance and applications of the area. Next, I review the existing algorithms, how they work, and their pros and cons. Finally, I go into more detail on co-occurrence matrix, which is the method I will be using.

Before I start designing the algorithm, I need to set up a system which can be used to measure the efficacy of my method. So, I created training and testing image sets and a software that can evaluate a texture analysis algorithm on these image sets.

Finally, I propose and implement a simple algorithm for the adjustment of one of the important parameters of the co-occurrence matrix, de offset vector. I use another image property, the energy spectrum, to choose the value.

At the end of the paper I evaluate my method, which doesn't give satisfactory results in its current state, but looks promising with further improvements.


Please sign in to download the files of this thesis.